leamsigc/shortsgenerator/facebook-research-template

Claude Code and agent skill by leamsigc/shortsgenerator

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Download the top 5 best-performing reels from a scraped Facebook page, transcribe them with Whisper, analyze the transcripts for structural patterns, and generate a reusable video template based on the best performer so the user can replicate or exceed its results.

Install

npx vskill@latest install leamsigc/shortsgenerator/facebook-research-template

SKILL.md

facebook-research-template — Video Template Extraction

Full pipeline: Download → Transcribe → Analyze → Template → Update REPORT.html

Prerequisites

  • openai-whisper installed (pip install openai-whisper)
  • ffmpeg installed
  • Chrome logged into Facebook (needed to access video pages)
  • Previous states completed (scrape + analyze must have run)

Pipeline Steps

Step 1: Download Top 5 Reels
bash
node .fb-research/scripts/download-reels.js <project-name>

The script navigates to each of the top 5 posts (from page-analytics.json top5), extracts the <video> src, and downloads to video-template/downloads/post-N.mp4. If src is a blob URL or download fails, fall back to Chrome DevTools manual capture of the network request that serves the video.

Step 2: Transcribe with Whisper
bash
bash .fb-research/scripts/transcribe.sh <project-name>

Processes every file in video-template/downloads/ with Whisper base model (Spanish) and writes .txt + .srt to video-template/transcripts/. Skips already-transcribed files. Adjust --model in the script for speed (tiny) or accuracy (large).

Step 3: Analyze Transcripts

For each of the 5 transcripts, extract:

  1. Hook (0-5s): Exact wording, timing, tone
  2. Context (5-15s): How the situation is established
  3. Body/Drama (15-40s): The narrative arc, key moments
  4. CTA (40-50s): Call to action format
  5. Pacing: Sentence length distribution, questions per video, emphasis patterns
  6. Vocabulary: Power words, repeated phrases, emotional triggers
  7. Emotional arc: How the video makes the viewer feel moment-by-moment

Compare all 5 to find:

  • What the best performer (#1) does differently
  • Common patterns across all top performers
  • What the worst performer (#5) is missing
Step 4: Generate Reusable Template

Create a fill-in-the-blank template based on the #1 post's structure.

The template includes:

  • Exact hook formula with slots: ¡[STAR NAME] + [DRAMATIC VERB] + [TEAM]! [EMOJI]
  • Scene-by-scene structure with timing
  • Vocabulary list (power words that work)
  • CTA formula
  • Production specs (duration, text style, audio)
Step 5: Update REPORT.html

Add a new section "🎬 Video Template" to REPORT.html showing:

  • The reusable template
  • The best performer breakdown
  • Comparison table of all 5 transcripts

Output Files

FileDescription
video-template/downloads/post-1.mp4 through post-5.mp4Downloaded reels
video-template/transcripts/post-1.txt through post-5.txtWhisper transcriptions
video-template/transcripts/post-1.srt through post-5.srtWhisper subtitles with timestamps
video-template/transcript-analysis.jsonStructured analysis of all 5 transcripts
video-template/VIDEO-TEMPLATE.mdReusable fill-in-the-blank video template
video-template/best-performer-breakdown.mdDeep analysis of the winner

Video Template Format

markdown
# Video Template: [Pattern Name]
**Based on**: [Post Title] — [views] views · [engagement] engagement
**Pattern**: [star+drama / upset-alert / curiosity-gap]
## 1. Hook Formula (0:00-0:03)
¡[STAR_NAME] [DRAMATIC_VERB] [TEAM/COUNTRY]! [EMOJI]
**Power words that work**:
-SALVA, ELIMINA, ROMPE, DESTROZA, HACE MAGIA
-INCREÍBLE, IMPOSIBLE, NADIE LO ESPERABA
-POLÉMICA, ROBO, INJUSTICIA, SOPRESA
## 2. Scene Structure
| Time | Section | Content | Emotion |
|------|---------|---------|---------|
| 0:00-0:03 | Hook | Bold text overlay + dramatic audio | Shock/Curiosity |
| 0:03-0:10 | Context | "Team was [situation]. But then..." | Tension |
| 0:10-0:30 | Drama | Key moment + reaction + analysis | Peak emotion |
| 0:30-0:45 | Opinion | "The real story is..." → controversial take | Debate |
| 0:45-0:50 | CTA | Question + subscribe prompt | Engagement |
## 3. Narrative Arc
1.**Pattern interrupt** (0s): Bold claim that stops scroll
2.**Context hook** (3s): "Everyone thought X, but Y happened"
3.**Escalation** (10s): Building the drama step by step
4.**Climax** (25s): The key moment described vividly
5.**Resolution** (35s): Your hot take/analysis
6.**Engagement bait** (45s): Question that splits the audience
## 4. Caption Formula
¡[HOOK]! [EMOJI]
[2-3 sentences: context + drama + opinion]
[CONTROVERSIAL QUESTION]
¿Crees que...? ¡Te leo en los comentarios! 👇
[hashtags: 2 viral + 2 niche + 1 branded]
## 5. Production Specs
-Duration: 45-50 seconds
-Text overlay: Bold upper-case, white with black stroke
-Audio: Match commentary + cinematic background music
-Captions: Styled (bold_outline or modern_glow)
-Transitions: Fast cuts (1-2 second clips)
-Color: High contrast, warm tones for excitement
## 6. Fill-in-the-Blanks Template
Copy this for each new video:

Hook: ¡[____] [____] [____]! [emoji] Context: [Team] estaba [situation]. Pero entonces... Drama: [Key moment] significó [impact] para [team]. Opinion: Lo que nadie dice es [controversial take]. CTA: ¿Crees que [debate question]? ¡Te leo! 👇

Notes

  • If download fails (Facebook blocks), generate the template from the available metadata + analysis data
  • Whisper model can be adjusted: tiny (fast) to large (accurate)
  • For Spanish content, base model works well; use large for accented speech

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
348
Repository forks
75
7-day trend score
29

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/leamsigc/shortsgenerator/facebook-research-template/badge)](https://verified-skill.com/skills/leamsigc/shortsgenerator/facebook-research-template)

Something wrong with this skill or its page? Tell us.